ruff/crates/red_knot_python_semantic/mdtest.py
David Peter 3235cd8019
[red-knot] Fix possible TOCTOU mistake in mdtest runner (#15673)
## Summary

Somehow, I managed to crash the `mdtest` runner today. I struggled to
reproduce this again to see if it's actually fixed (even with an
artificial `sleep` between the two `cargo test` invocations), but the
original backtrace clearly showed that this is where the problem
originated from. And it seems like a clear TOCTOU problem.
2025-01-22 15:24:25 +00:00

220 lines
7.4 KiB
Python

"""A runner for Markdown-based tests for Red Knot"""
# /// script
# requires-python = ">=3.11"
# dependencies = [
# "rich",
# "watchfiles",
# ]
# ///
from __future__ import annotations
import json
import os
import subprocess
from pathlib import Path
from typing import Final, Literal, Never, assert_never
from rich.console import Console
from watchfiles import Change, watch
CRATE_NAME: Final = "red_knot_python_semantic"
CRATE_ROOT: Final = Path(__file__).resolve().parent
MDTEST_DIR: Final = CRATE_ROOT / "resources" / "mdtest"
class MDTestRunner:
mdtest_executable: Path | None
console: Console
def __init__(self) -> None:
self.mdtest_executable = None
self.console = Console()
def _run_cargo_test(self, *, message_format: Literal["human", "json"]) -> str:
return subprocess.check_output(
[
"cargo",
"test",
"--package",
CRATE_NAME,
"--no-run",
"--color=always",
"--message-format",
message_format,
],
cwd=CRATE_ROOT,
env=dict(os.environ, CLI_COLOR="1"),
stderr=subprocess.STDOUT,
text=True,
)
def _recompile_tests(
self, status_message: str, *, message_on_success: bool = True
) -> bool:
with self.console.status(status_message):
# Run it with 'human' format in case there are errors:
try:
self._run_cargo_test(message_format="human")
except subprocess.CalledProcessError as e:
print(e.output)
return False
# Run it again with 'json' format to find the mdtest executable:
try:
json_output = self._run_cargo_test(message_format="json")
except subprocess.CalledProcessError as _:
# `cargo test` can still fail if something changed in between the two runs.
# Here we don't have a human-readable output, so just show a generic message:
self.console.print("[red]Error[/red]: Failed to compile tests")
return False
if json_output:
self._get_executable_path_from_json(json_output)
if message_on_success:
self.console.print("[dim]Tests compiled successfully[/dim]")
return True
def _get_executable_path_from_json(self, json_output: str) -> None:
for json_line in json_output.splitlines():
try:
data = json.loads(json_line)
except json.JSONDecodeError:
continue
if data.get("target", {}).get("name") == "mdtest":
self.mdtest_executable = Path(data["executable"])
break
else:
raise RuntimeError(
"Could not find mdtest executable after successful compilation"
)
def _run_mdtest(
self, arguments: list[str] | None = None, *, capture_output: bool = False
) -> subprocess.CompletedProcess:
assert self.mdtest_executable is not None
arguments = arguments or []
return subprocess.run(
[self.mdtest_executable, *arguments],
cwd=CRATE_ROOT,
env=dict(os.environ, CLICOLOR_FORCE="1"),
capture_output=capture_output,
text=True,
check=False,
)
def _run_mdtests_for_file(self, markdown_file: Path) -> None:
path_mangled = (
markdown_file.as_posix()
.replace("/", "_")
.replace("-", "_")
.removesuffix(".md")
)
test_name = f"mdtest__{path_mangled}"
output = self._run_mdtest(["--exact", test_name], capture_output=True)
if output.returncode == 0:
if "running 0 tests\n" in output.stdout:
self.console.log(
f"[yellow]Warning[/yellow]: No tests were executed with filter '{test_name}'"
)
else:
self.console.print(
f"Test for [bold green]{markdown_file}[/bold green] succeeded"
)
else:
self.console.print()
self.console.rule(
f"Test for [bold red]{markdown_file}[/bold red] failed",
style="gray",
)
self._print_trimmed_cargo_test_output(
output.stdout + output.stderr, test_name
)
def _print_trimmed_cargo_test_output(self, output: str, test_name: str) -> None:
# Skip 'cargo test' boilerplate at the beginning:
lines = output.splitlines()
start_index = 0
for i, line in enumerate(lines):
if f"{test_name} stdout" in line:
start_index = i
break
for line in lines[start_index + 1 :]:
if "MDTEST_TEST_FILTER" in line:
continue
if line.strip() == "-" * 50:
# Skip 'cargo test' boilerplate at the end
break
print(line)
def watch(self) -> Never:
self._recompile_tests("Compiling tests...", message_on_success=False)
self.console.print("[dim]Ready to watch for changes...[/dim]")
for changes in watch(CRATE_ROOT):
new_md_files = set()
changed_md_files = set()
rust_code_has_changed = False
for change, path_str in changes:
path = Path(path_str)
if path.suffix == ".rs":
rust_code_has_changed = True
continue
if path.suffix != ".md":
continue
try:
relative_path = Path(path).relative_to(MDTEST_DIR)
except ValueError:
continue
match change:
case Change.added:
# When saving a file, some editors (looking at you, Vim) might first
# save the file with a temporary name (e.g. `file.md~`) and then rename
# it to the final name. This creates a `deleted` and `added` change.
# We treat those files as `changed` here.
if (Change.deleted, path_str) in changes:
changed_md_files.add(relative_path)
else:
new_md_files.add(relative_path)
case Change.modified:
changed_md_files.add(relative_path)
case Change.deleted:
# No need to do anything when a Markdown test is deleted
pass
case _ as unreachable:
assert_never(unreachable)
if rust_code_has_changed:
if self._recompile_tests("Rust code has changed, recompiling tests..."):
self._run_mdtest()
elif new_md_files:
files = " ".join(file.as_posix() for file in new_md_files)
self._recompile_tests(
f"New Markdown test [yellow]{files}[/yellow] detected, recompiling tests..."
)
for path in new_md_files | changed_md_files:
self._run_mdtests_for_file(path)
def main() -> None:
try:
runner = MDTestRunner()
runner.watch()
except KeyboardInterrupt:
print()
if __name__ == "__main__":
main()